start work rewriting timeline cache type

This commit is contained in:
kim 2024-12-30 17:12:55 +00:00
commit f4b4a696f2
23 changed files with 1792 additions and 861 deletions

View file

@ -20,15 +20,13 @@ package bundb
import (
"context"
"errors"
"fmt"
"slices"
"time"
"github.com/superseriousbusiness/gotosocial/internal/db"
"github.com/superseriousbusiness/gotosocial/internal/gtscontext"
"github.com/superseriousbusiness/gotosocial/internal/gtserror"
"github.com/superseriousbusiness/gotosocial/internal/gtsmodel"
"github.com/superseriousbusiness/gotosocial/internal/id"
"github.com/superseriousbusiness/gotosocial/internal/paging"
"github.com/superseriousbusiness/gotosocial/internal/state"
"github.com/uptrace/bun"
)
@ -38,161 +36,97 @@ type timelineDB struct {
state *state.State
}
func (t *timelineDB) GetHomeTimeline(ctx context.Context, accountID string, maxID string, sinceID string, minID string, limit int, local bool) ([]*gtsmodel.Status, error) {
// Ensure reasonable
if limit < 0 {
limit = 0
}
func (t *timelineDB) GetHomeTimeline(ctx context.Context, accountID string, page *paging.Page) ([]*gtsmodel.Status, error) {
return loadStatusTimelinePage(ctx, t.db, t.state,
// Make educated guess for slice size
var (
statusIDs = make([]string, 0, limit)
frontToBack = true
)
// Paging
// params.
page,
// As this is the home timeline, it should be
// populated by statuses from accounts followed
// by accountID, and posts from accountID itself.
//
// So, begin by seeing who accountID follows.
// It should be a little cheaper to do this in
// a separate query like this, rather than using
// a join, since followIDs are cached in memory.
follows, err := t.state.DB.GetAccountFollows(
gtscontext.SetBarebones(ctx),
accountID,
nil, // select all
)
if err != nil && !errors.Is(err, db.ErrNoEntries) {
return nil, gtserror.Newf("db error getting follows for account %s: %w", accountID, err)
}
// The actual meat of the home-timeline query, outside
// of any paging parameters that selects by followings.
func(q *bun.SelectQuery) (*bun.SelectQuery, error) {
// To take account of exclusive lists, get all of
// this account's lists, so we can filter out follows
// that are in contained in exclusive lists.
lists, err := t.state.DB.GetListsByAccountID(ctx, accountID)
if err != nil && !errors.Is(err, db.ErrNoEntries) {
return nil, gtserror.Newf("db error getting lists for account %s: %w", accountID, err)
}
// Index all follow IDs that fall in exclusive lists.
ignoreFollowIDs := make(map[string]struct{})
for _, list := range lists {
if !*list.Exclusive {
// Not exclusive,
// we don't care.
continue
}
// Fetch all follow IDs of the entries ccontained in this list.
listFollowIDs, err := t.state.DB.GetFollowIDsInList(ctx, list.ID, nil)
if err != nil && !errors.Is(err, db.ErrNoEntries) {
return nil, gtserror.Newf("db error getting list entry follow ids: %w", err)
}
// Exclusive list, index all its follow IDs.
for _, followID := range listFollowIDs {
ignoreFollowIDs[followID] = struct{}{}
}
}
// Extract just the accountID from each follow,
// ignoring follows that are in exclusive lists.
targetAccountIDs := make([]string, 0, len(follows)+1)
for _, f := range follows {
_, ignore := ignoreFollowIDs[f.ID]
if !ignore {
targetAccountIDs = append(
targetAccountIDs,
f.TargetAccountID,
// As this is the home timeline, it should be
// populated by statuses from accounts followed
// by accountID, and posts from accountID itself.
//
// So, begin by seeing who accountID follows.
// It should be a little cheaper to do this in
// a separate query like this, rather than using
// a join, since followIDs are cached in memory.
follows, err := t.state.DB.GetAccountFollows(
gtscontext.SetBarebones(ctx),
accountID,
nil, // select all
)
}
}
if err != nil && !errors.Is(err, db.ErrNoEntries) {
return nil, gtserror.Newf("db error getting follows for account %s: %w", accountID, err)
}
// Add accountID itself as a pseudo follow so that
// accountID can see its own posts in the timeline.
targetAccountIDs = append(targetAccountIDs, accountID)
// To take account of exclusive lists, get all of
// this account's lists, so we can filter out follows
// that are in contained in exclusive lists.
lists, err := t.state.DB.GetListsByAccountID(ctx, accountID)
if err != nil && !errors.Is(err, db.ErrNoEntries) {
return nil, gtserror.Newf("db error getting lists for account %s: %w", accountID, err)
}
// Now start building the database query.
q := t.db.
NewSelect().
TableExpr("? AS ?", bun.Ident("statuses"), bun.Ident("status")).
// Select only IDs from table
Column("status.id")
// Index all follow IDs that fall in exclusive lists.
ignoreFollowIDs := make(map[string]struct{})
for _, list := range lists {
if !*list.Exclusive {
// Not exclusive,
// we don't care.
continue
}
if maxID == "" || maxID >= id.Highest {
const future = 24 * time.Hour
// Fetch all follow IDs of the entries ccontained in this list.
listFollowIDs, err := t.state.DB.GetFollowIDsInList(ctx, list.ID, nil)
if err != nil && !errors.Is(err, db.ErrNoEntries) {
return nil, gtserror.Newf("db error getting list entry follow ids: %w", err)
}
// don't return statuses more than 24hr in the future
maxID = id.NewULIDFromTime(time.Now().Add(future))
}
// Exclusive list, index all its follow IDs.
for _, followID := range listFollowIDs {
ignoreFollowIDs[followID] = struct{}{}
}
}
// return only statuses LOWER (ie., older) than maxID
q = q.Where("? < ?", bun.Ident("status.id"), maxID)
// Extract just the accountID from each follow,
// ignoring follows that are in exclusive lists.
targetAccountIDs := make([]string, 0, len(follows)+1)
for _, f := range follows {
_, ignore := ignoreFollowIDs[f.ID]
if !ignore {
targetAccountIDs = append(
targetAccountIDs,
f.TargetAccountID,
)
}
}
if sinceID != "" {
// return only statuses HIGHER (ie., newer) than sinceID
q = q.Where("? > ?", bun.Ident("status.id"), sinceID)
}
// Add accountID itself as a pseudo follow so that
// accountID can see its own posts in the timeline.
targetAccountIDs = append(targetAccountIDs, accountID)
if minID != "" {
// return only statuses HIGHER (ie., newer) than minID
q = q.Where("? > ?", bun.Ident("status.id"), minID)
// Select only statuses authored by
// accounts with IDs in the slice.
q = q.Where(
"? IN (?)",
bun.Ident("account_id"),
bun.In(targetAccountIDs),
)
// page up
frontToBack = false
}
// Only include statuses that aren't pending approval.
q = q.Where("NOT ? = ?", bun.Ident("pending_approval"), true)
if local {
// return only statuses posted by local account havers
q = q.Where("? = ?", bun.Ident("status.local"), local)
}
// Select only statuses authored by
// accounts with IDs in the slice.
q = q.Where(
"? IN (?)",
bun.Ident("status.account_id"),
bun.In(targetAccountIDs),
return q, nil
},
)
// Only include statuses that aren't pending approval.
q = q.Where("NOT ? = ?", bun.Ident("status.pending_approval"), true)
if limit > 0 {
// limit amount of statuses returned
q = q.Limit(limit)
}
if frontToBack {
// Page down.
q = q.Order("status.id DESC")
} else {
// Page up.
q = q.Order("status.id ASC")
}
if err := q.Scan(ctx, &statusIDs); err != nil {
return nil, err
}
if len(statusIDs) == 0 {
return nil, nil
}
// If we're paging up, we still want statuses
// to be sorted by ID desc, so reverse ids slice.
// https://zchee.github.io/golang-wiki/SliceTricks/#reversing
if !frontToBack {
for l, r := 0, len(statusIDs)-1; l < r; l, r = l+1, r-1 {
statusIDs[l], statusIDs[r] = statusIDs[r], statusIDs[l]
}
}
// Return status IDs loaded from cache + db.
return t.state.DB.GetStatusesByIDs(ctx, statusIDs)
}
<<<<<<< HEAD
func (t *timelineDB) GetPublicTimeline(
ctx context.Context,
maxID string,
@ -289,6 +223,28 @@ func (t *timelineDB) GetPublicTimeline(
// Return status IDs loaded from cache + db.
return t.state.DB.GetStatusesByIDs(ctx, statusIDs)
=======
func (t *timelineDB) GetPublicTimeline(ctx context.Context, page *paging.Page) ([]*gtsmodel.Status, error) {
return loadStatusTimelinePage(ctx, t.db, t.state,
// Paging
// params.
page,
func(q *bun.SelectQuery) (*bun.SelectQuery, error) {
// Public only.
q = q.Where("? = ?", bun.Ident("visibility"), gtsmodel.VisibilityPublic)
// Ignore boosts.
q = q.Where("? IS NULL", bun.Ident("boost_of_id"))
// Only include statuses that aren't pending approval.
q = q.Where("NOT ? = ?", bun.Ident("pending_approval"), true)
return q, nil
},
)
>>>>>>> 6f0abe7fb (start work rewriting timeline cache type)
}
func (t *timelineDB) getLocalTimeline(
@ -378,6 +334,7 @@ func (t *timelineDB) getLocalTimeline(
// TODO optimize this query and the logic here, because it's slow as balls -- it takes like a literal second to return with a limit of 20!
// It might be worth serving it through a timeline instead of raw DB queries, like we do for Home feeds.
func (t *timelineDB) GetFavedTimeline(ctx context.Context, accountID string, maxID string, minID string, limit int) ([]*gtsmodel.Status, string, string, error) {
// Ensure reasonable
if limit < 0 {
limit = 0
@ -442,205 +399,138 @@ func (t *timelineDB) GetFavedTimeline(ctx context.Context, accountID string, max
return statuses, nextMaxID, prevMinID, nil
}
func (t *timelineDB) GetListTimeline(
func (t *timelineDB) GetListTimeline(ctx context.Context, listID string, page *paging.Page) ([]*gtsmodel.Status, error) {
return loadStatusTimelinePage(ctx, t.db, t.state,
// Paging
// params.
page,
// The actual meat of the list-timeline query, outside
// of any paging parameters, it selects by list entries.
func(q *bun.SelectQuery) (*bun.SelectQuery, error) {
// Fetch all follow IDs contained in list from DB.
followIDs, err := t.state.DB.GetFollowIDsInList(
ctx, listID, nil,
)
if err != nil {
return nil, gtserror.Newf("error getting follows in list: %w", err)
}
// Select target account
// IDs from list follows.
subQ := t.db.NewSelect().
TableExpr("? AS ?", bun.Ident("follows"), bun.Ident("follow")).
Column("follow.target_account_id").
Where("? IN (?)", bun.Ident("follow.id"), bun.In(followIDs))
q = q.Where("? IN (?)", bun.Ident("status.account_id"), subQ)
// Only include statuses that aren't pending approval.
q = q.Where("NOT ? = ?", bun.Ident("pending_approval"), true)
return q, nil
},
)
}
func (t *timelineDB) GetTagTimeline(ctx context.Context, tagID string, page *paging.Page) ([]*gtsmodel.Status, error) {
return loadStatusTimelinePage(ctx, t.db, t.state,
// Paging
// params.
page,
// The actual meat of the list-timeline query, outside of any
// paging params, selects by status tags with public visibility.
func(q *bun.SelectQuery) (*bun.SelectQuery, error) {
// ...
q = q.Join(
"INNER JOIN ? ON ? = ?",
bun.Ident("status_to_tags"),
bun.Ident("status.id"), bun.Ident("status_to_tags.status_id"),
)
// This tag only.
q = q.Where("? = ?", bun.Ident("status_to_tags.tag_id"), tagID)
// Public only.
q = q.Where("? = ?", bun.Ident("status.visibility"), gtsmodel.VisibilityPublic)
return q, nil
},
)
}
func loadStatusTimelinePage(
ctx context.Context,
listID string,
maxID string,
sinceID string,
minID string,
limit int,
) ([]*gtsmodel.Status, error) {
// Ensure reasonable
if limit < 0 {
limit = 0
db *bun.DB,
state *state.State,
page *paging.Page,
query func(*bun.SelectQuery) (*bun.SelectQuery, error),
) (
[]*gtsmodel.Status,
error,
) {
// Extract page params.
minID := page.Min.Value
maxID := page.Max.Value
limit := page.Limit
order := page.Order()
// Pre-allocate slice of IDs as dest.
statusIDs := make([]string, 0, limit)
// Now start building the database query.
//
// Select the following:
// - status ID
q := db.NewSelect().
Table("statuses").
Column("id")
if maxID != "" {
// Set a maximum ID boundary if was given.
q = q.Where("? < ?", bun.Ident("id"), maxID)
}
// Make educated guess for slice size
var (
statusIDs = make([]string, 0, limit)
frontToBack = true
)
if minID != "" {
// Set a minimum ID boundary if was given.
q = q.Where("? > ?", bun.Ident("id"), minID)
}
// Fetch all follow IDs contained in list from DB.
followIDs, err := t.state.DB.GetFollowIDsInList(
ctx, listID, nil,
)
// Append caller
// query details.
q, err := query(q)
if err != nil {
return nil, fmt.Errorf("error getting follows in list: %w", err)
return nil, err
}
// If there's no list follows we can't
// possibly return anything for this list.
if len(followIDs) == 0 {
return make([]*gtsmodel.Status, 0), nil
// Set ordering.
switch order {
case paging.OrderAscending:
q = q.OrderExpr("? ASC", bun.Ident("id"))
case paging.OrderDescending:
q = q.OrderExpr("? DESC", bun.Ident("id"))
}
// Select target account IDs from follows.
subQ := t.db.
NewSelect().
TableExpr("? AS ?", bun.Ident("follows"), bun.Ident("follow")).
Column("follow.target_account_id").
Where("? IN (?)", bun.Ident("follow.id"), bun.In(followIDs))
// Select only status IDs created
// by one of the followed accounts.
q := t.db.
NewSelect().
TableExpr("? AS ?", bun.Ident("statuses"), bun.Ident("status")).
// Select only IDs from table
Column("status.id").
Where("? IN (?)", bun.Ident("status.account_id"), subQ)
if maxID == "" || maxID >= id.Highest {
const future = 24 * time.Hour
// don't return statuses more than 24hr in the future
maxID = id.NewULIDFromTime(time.Now().Add(future))
}
// return only statuses LOWER (ie., older) than maxID
q = q.Where("? < ?", bun.Ident("status.id"), maxID)
if sinceID != "" {
// return only statuses HIGHER (ie., newer) than sinceID
q = q.Where("? > ?", bun.Ident("status.id"), sinceID)
}
if minID != "" {
// return only statuses HIGHER (ie., newer) than minID
q = q.Where("? > ?", bun.Ident("status.id"), minID)
// page up
frontToBack = false
}
// Only include statuses that aren't pending approval.
q = q.Where("NOT ? = ?", bun.Ident("status.pending_approval"), true)
if limit > 0 {
// limit amount of statuses returned
q = q.Limit(limit)
}
if frontToBack {
// Page down.
q = q.Order("status.id DESC")
} else {
// Page up.
q = q.Order("status.id ASC")
}
// A limit should always
// be supplied for this.
q = q.Limit(limit)
// Finally, perform query into status ID slice.
if err := q.Scan(ctx, &statusIDs); err != nil {
return nil, err
}
if len(statusIDs) == 0 {
return nil, nil
// The order we return from the database and
// timeline caches differs depending on ordering,
// but the caller always expected DESCENDING.
if page.GetOrder() == paging.OrderAscending {
slices.Reverse(statusIDs)
}
// If we're paging up, we still want statuses
// to be sorted by ID desc, so reverse ids slice.
// https://zchee.github.io/golang-wiki/SliceTricks/#reversing
if !frontToBack {
for l, r := 0, len(statusIDs)-1; l < r; l, r = l+1, r-1 {
statusIDs[l], statusIDs[r] = statusIDs[r], statusIDs[l]
}
}
// Return status IDs loaded from cache + db.
return t.state.DB.GetStatusesByIDs(ctx, statusIDs)
}
func (t *timelineDB) GetTagTimeline(
ctx context.Context,
tagID string,
maxID string,
sinceID string,
minID string,
limit int,
) ([]*gtsmodel.Status, error) {
// Ensure reasonable
if limit < 0 {
limit = 0
}
// Make educated guess for slice size
var (
statusIDs = make([]string, 0, limit)
frontToBack = true
)
q := t.db.
NewSelect().
TableExpr("? AS ?", bun.Ident("status_to_tags"), bun.Ident("status_to_tag")).
Column("status_to_tag.status_id").
// Join with statuses for filtering.
Join(
"INNER JOIN ? AS ? ON ? = ?",
bun.Ident("statuses"), bun.Ident("status"),
bun.Ident("status.id"), bun.Ident("status_to_tag.status_id"),
).
// Public only.
Where("? = ?", bun.Ident("status.visibility"), gtsmodel.VisibilityPublic).
// This tag only.
Where("? = ?", bun.Ident("status_to_tag.tag_id"), tagID)
if maxID == "" || maxID >= id.Highest {
const future = 24 * time.Hour
// don't return statuses more than 24hr in the future
maxID = id.NewULIDFromTime(time.Now().Add(future))
}
// return only statuses LOWER (ie., older) than maxID
q = q.Where("? < ?", bun.Ident("status_to_tag.status_id"), maxID)
if sinceID != "" {
// return only statuses HIGHER (ie., newer) than sinceID
q = q.Where("? > ?", bun.Ident("status_to_tag.status_id"), sinceID)
}
if minID != "" {
// return only statuses HIGHER (ie., newer) than minID
q = q.Where("? > ?", bun.Ident("status_to_tag.status_id"), minID)
// page up
frontToBack = false
}
// Only include statuses that aren't pending approval.
q = q.Where("NOT ? = ?", bun.Ident("status.pending_approval"), true)
if limit > 0 {
// limit amount of statuses returned
q = q.Limit(limit)
}
if frontToBack {
// Page down.
q = q.Order("status_to_tag.status_id DESC")
} else {
// Page up.
q = q.Order("status_to_tag.status_id ASC")
}
if err := q.Scan(ctx, &statusIDs); err != nil {
return nil, err
}
if len(statusIDs) == 0 {
return nil, nil
}
// If we're paging up, we still want statuses
// to be sorted by ID desc, so reverse ids slice.
// https://zchee.github.io/golang-wiki/SliceTricks/#reversing
if !frontToBack {
for l, r := 0, len(statusIDs)-1; l < r; l, r = l+1, r-1 {
statusIDs[l], statusIDs[r] = statusIDs[r], statusIDs[l]
}
}
// Return status IDs loaded from cache + db.
return t.state.DB.GetStatusesByIDs(ctx, statusIDs)
// Fetch statuses from DB / cache with given IDs.
return state.DB.GetStatusesByIDs(ctx, statusIDs)
}